TypeError: __deepcopy__() takes 1 positional argument but 2 were given

[omerfguzel]

Hi,
I'm getting this error when I run the ensemble_eval function. Is there any fix suggestions?
TypeError: deepcopy() takes 1 positional argument but 2 were given

[h3ik0th]

Hi,
after some experimenting I found that the class RegressionEnsembleModel in Darts has apparently been modified. The argument "forecasting_models" no longer accepts the list "models" of instantiated models.
Below, I commented out the syntax that worked in October and inserted a new line below. In argument "forecasting_models", I list the forecast methods explicitly, one by one.

old:
# ensemble_model = RegressionEnsembleModel(forecasting_models=models, regression_train_n_points=n_train)

new:
ensemble_model = RegressionEnsembleModel(
forecasting_models=[ Theta(best_theta), Prophet(), AutoARIMA(), NaiveDrift() ],
regression_train_n_points=12)

I still need to look if there is any workaround, so that RegressionEnsembleModel accepts instantiated forecast models. The Darts docu doesn't mention anything.

[h3ik0th]

Revised Juypter notebook uploaded: the new code line I mentioned above prevents the runtime error

[omerfguzel]

Thank you so much sir, solved my problem.

[dennisbader]

@h3ik0th and @omerfguzel, thanks for that. We will improve the error message, see Darts issue 724.

When creating a model object all parameters are saved.
Deepcopying fitted/trained forecasting_models with EnsembleModels fails.

You can reset your models with models_untrained = [m.untrained_model() for m in models]